1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * IPv4 Forwarding Information Base: semantics.
9 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
12 #include <linux/uaccess.h>
13 #include <linux/bitops.h>
14 #include <linux/types.h>
15 #include <linux/kernel.h>
16 #include <linux/jiffies.h>
18 #include <linux/string.h>
19 #include <linux/socket.h>
20 #include <linux/sockios.h>
21 #include <linux/errno.h>
23 #include <linux/inet.h>
24 #include <linux/inetdevice.h>
25 #include <linux/netdevice.h>
26 #include <linux/if_arp.h>
27 #include <linux/proc_fs.h>
28 #include <linux/skbuff.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/netlink.h>
35 #include <net/protocol.h>
36 #include <net/route.h>
39 #include <net/ip_fib.h>
40 #include <net/ip6_fib.h>
41 #include <net/nexthop.h>
42 #include <net/netlink.h>
44 #include <net/lwtunnel.h>
45 #include <net/fib_notifier.h>
46 #include <net/addrconf.h>
48 #include "fib_lookup.h"
50 static DEFINE_SPINLOCK(fib_info_lock);
51 static struct hlist_head *fib_info_hash;
52 static struct hlist_head *fib_info_laddrhash;
53 static unsigned int fib_info_hash_size;
54 static unsigned int fib_info_cnt;
56 #define DEVINDEX_HASHBITS 8
57 #define DEVINDEX_HASHSIZE (1U << DEVINDEX_HASHBITS)
58 static struct hlist_head fib_info_devhash[DEVINDEX_HASHSIZE];
60 /* for_nexthops and change_nexthops only used when nexthop object
61 * is not set in a fib_info. The logic within can reference fib_nh.
63 #ifdef CONFIG_IP_ROUTE_MULTIPATH
65 #define for_nexthops(fi) { \
66 int nhsel; const struct fib_nh *nh; \
67 for (nhsel = 0, nh = (fi)->fib_nh; \
68 nhsel < fib_info_num_path((fi)); \
71 #define change_nexthops(fi) { \
72 int nhsel; struct fib_nh *nexthop_nh; \
73 for (nhsel = 0, nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
74 nhsel < fib_info_num_path((fi)); \
75 nexthop_nh++, nhsel++)
77 #else /* CONFIG_IP_ROUTE_MULTIPATH */
79 /* Hope, that gcc will optimize it to get rid of dummy loop */
81 #define for_nexthops(fi) { \
82 int nhsel; const struct fib_nh *nh = (fi)->fib_nh; \
83 for (nhsel = 0; nhsel < 1; nhsel++)
85 #define change_nexthops(fi) { \
87 struct fib_nh *nexthop_nh = (struct fib_nh *)((fi)->fib_nh); \
88 for (nhsel = 0; nhsel < 1; nhsel++)
90 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
92 #define endfor_nexthops(fi) }
95 const struct fib_prop fib_props[RTN_MAX + 1] = {
98 .scope = RT_SCOPE_NOWHERE,
102 .scope = RT_SCOPE_UNIVERSE,
106 .scope = RT_SCOPE_HOST,
110 .scope = RT_SCOPE_LINK,
114 .scope = RT_SCOPE_LINK,
118 .scope = RT_SCOPE_UNIVERSE,
122 .scope = RT_SCOPE_UNIVERSE,
124 [RTN_UNREACHABLE] = {
125 .error = -EHOSTUNREACH,
126 .scope = RT_SCOPE_UNIVERSE,
130 .scope = RT_SCOPE_UNIVERSE,
134 .scope = RT_SCOPE_UNIVERSE,
138 .scope = RT_SCOPE_NOWHERE,
142 .scope = RT_SCOPE_NOWHERE,
146 static void rt_fibinfo_free(struct rtable __rcu **rtp)
148 struct rtable *rt = rcu_dereference_protected(*rtp, 1);
153 /* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
154 * because we waited an RCU grace period before calling
155 * free_fib_info_rcu()
158 dst_dev_put(&rt->dst);
159 dst_release_immediate(&rt->dst);
162 static void free_nh_exceptions(struct fib_nh_common *nhc)
164 struct fnhe_hash_bucket *hash;
167 hash = rcu_dereference_protected(nhc->nhc_exceptions, 1);
170 for (i = 0; i < FNHE_HASH_SIZE; i++) {
171 struct fib_nh_exception *fnhe;
173 fnhe = rcu_dereference_protected(hash[i].chain, 1);
175 struct fib_nh_exception *next;
177 next = rcu_dereference_protected(fnhe->fnhe_next, 1);
179 rt_fibinfo_free(&fnhe->fnhe_rth_input);
180 rt_fibinfo_free(&fnhe->fnhe_rth_output);
190 static void rt_fibinfo_free_cpus(struct rtable __rcu * __percpu *rtp)
197 for_each_possible_cpu(cpu) {
200 rt = rcu_dereference_protected(*per_cpu_ptr(rtp, cpu), 1);
202 dst_dev_put(&rt->dst);
203 dst_release_immediate(&rt->dst);
209 void fib_nh_common_release(struct fib_nh_common *nhc)
212 dev_put(nhc->nhc_dev);
214 lwtstate_put(nhc->nhc_lwtstate);
215 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
216 rt_fibinfo_free(&nhc->nhc_rth_input);
217 free_nh_exceptions(nhc);
219 EXPORT_SYMBOL_GPL(fib_nh_common_release);
221 void fib_nh_release(struct net *net, struct fib_nh *fib_nh)
223 #ifdef CONFIG_IP_ROUTE_CLASSID
224 if (fib_nh->nh_tclassid)
225 net->ipv4.fib_num_tclassid_users--;
227 fib_nh_common_release(&fib_nh->nh_common);
230 /* Release a nexthop info record */
231 static void free_fib_info_rcu(struct rcu_head *head)
233 struct fib_info *fi = container_of(head, struct fib_info, rcu);
238 change_nexthops(fi) {
239 fib_nh_release(fi->fib_net, nexthop_nh);
240 } endfor_nexthops(fi);
243 ip_fib_metrics_put(fi->fib_metrics);
248 void free_fib_info(struct fib_info *fi)
250 if (fi->fib_dead == 0) {
251 pr_warn("Freeing alive fib_info %p\n", fi);
256 call_rcu(&fi->rcu, free_fib_info_rcu);
258 EXPORT_SYMBOL_GPL(free_fib_info);
260 void fib_release_info(struct fib_info *fi)
262 spin_lock_bh(&fib_info_lock);
263 if (fi && --fi->fib_treeref == 0) {
264 hlist_del(&fi->fib_hash);
266 hlist_del(&fi->fib_lhash);
268 list_del(&fi->nh_list);
270 change_nexthops(fi) {
271 if (!nexthop_nh->fib_nh_dev)
273 hlist_del(&nexthop_nh->nh_hash);
274 } endfor_nexthops(fi)
279 spin_unlock_bh(&fib_info_lock);
282 static inline int nh_comp(struct fib_info *fi, struct fib_info *ofi)
284 const struct fib_nh *onh;
286 if (fi->nh || ofi->nh)
287 return nexthop_cmp(fi->nh, ofi->nh) ? 0 : -1;
289 if (ofi->fib_nhs == 0)
293 onh = fib_info_nh(ofi, nhsel);
295 if (nh->fib_nh_oif != onh->fib_nh_oif ||
296 nh->fib_nh_gw_family != onh->fib_nh_gw_family ||
297 nh->fib_nh_scope != onh->fib_nh_scope ||
298 #ifdef CONFIG_IP_ROUTE_MULTIPATH
299 nh->fib_nh_weight != onh->fib_nh_weight ||
301 #ifdef CONFIG_IP_ROUTE_CLASSID
302 nh->nh_tclassid != onh->nh_tclassid ||
304 lwtunnel_cmp_encap(nh->fib_nh_lws, onh->fib_nh_lws) ||
305 ((nh->fib_nh_flags ^ onh->fib_nh_flags) & ~RTNH_COMPARE_MASK))
308 if (nh->fib_nh_gw_family == AF_INET &&
309 nh->fib_nh_gw4 != onh->fib_nh_gw4)
312 if (nh->fib_nh_gw_family == AF_INET6 &&
313 ipv6_addr_cmp(&nh->fib_nh_gw6, &onh->fib_nh_gw6))
315 } endfor_nexthops(fi);
319 static inline unsigned int fib_devindex_hashfn(unsigned int val)
321 unsigned int mask = DEVINDEX_HASHSIZE - 1;
324 (val >> DEVINDEX_HASHBITS) ^
325 (val >> (DEVINDEX_HASHBITS * 2))) & mask;
328 static unsigned int fib_info_hashfn_1(int init_val, u8 protocol, u8 scope,
329 u32 prefsrc, u32 priority)
331 unsigned int val = init_val;
333 val ^= (protocol << 8) | scope;
340 static unsigned int fib_info_hashfn_result(unsigned int val)
342 unsigned int mask = (fib_info_hash_size - 1);
344 return (val ^ (val >> 7) ^ (val >> 12)) & mask;
347 static inline unsigned int fib_info_hashfn(struct fib_info *fi)
351 val = fib_info_hashfn_1(fi->fib_nhs, fi->fib_protocol,
352 fi->fib_scope, (__force u32)fi->fib_prefsrc,
356 val ^= fib_devindex_hashfn(fi->nh->id);
359 val ^= fib_devindex_hashfn(nh->fib_nh_oif);
360 } endfor_nexthops(fi)
363 return fib_info_hashfn_result(val);
366 /* no metrics, only nexthop id */
367 static struct fib_info *fib_find_info_nh(struct net *net,
368 const struct fib_config *cfg)
370 struct hlist_head *head;
374 hash = fib_info_hashfn_1(fib_devindex_hashfn(cfg->fc_nh_id),
375 cfg->fc_protocol, cfg->fc_scope,
376 (__force u32)cfg->fc_prefsrc,
378 hash = fib_info_hashfn_result(hash);
379 head = &fib_info_hash[hash];
381 hlist_for_each_entry(fi, head, fib_hash) {
382 if (!net_eq(fi->fib_net, net))
384 if (!fi->nh || fi->nh->id != cfg->fc_nh_id)
386 if (cfg->fc_protocol == fi->fib_protocol &&
387 cfg->fc_scope == fi->fib_scope &&
388 cfg->fc_prefsrc == fi->fib_prefsrc &&
389 cfg->fc_priority == fi->fib_priority &&
390 cfg->fc_type == fi->fib_type &&
391 cfg->fc_table == fi->fib_tb_id &&
392 !((cfg->fc_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK))
399 static struct fib_info *fib_find_info(struct fib_info *nfi)
401 struct hlist_head *head;
405 hash = fib_info_hashfn(nfi);
406 head = &fib_info_hash[hash];
408 hlist_for_each_entry(fi, head, fib_hash) {
409 if (!net_eq(fi->fib_net, nfi->fib_net))
411 if (fi->fib_nhs != nfi->fib_nhs)
413 if (nfi->fib_protocol == fi->fib_protocol &&
414 nfi->fib_scope == fi->fib_scope &&
415 nfi->fib_prefsrc == fi->fib_prefsrc &&
416 nfi->fib_priority == fi->fib_priority &&
417 nfi->fib_type == fi->fib_type &&
418 memcmp(nfi->fib_metrics, fi->fib_metrics,
419 sizeof(u32) * RTAX_MAX) == 0 &&
420 !((nfi->fib_flags ^ fi->fib_flags) & ~RTNH_COMPARE_MASK) &&
421 nh_comp(fi, nfi) == 0)
428 /* Check, that the gateway is already configured.
429 * Used only by redirect accept routine.
431 int ip_fib_check_default(__be32 gw, struct net_device *dev)
433 struct hlist_head *head;
437 spin_lock(&fib_info_lock);
439 hash = fib_devindex_hashfn(dev->ifindex);
440 head = &fib_info_devhash[hash];
441 hlist_for_each_entry(nh, head, nh_hash) {
442 if (nh->fib_nh_dev == dev &&
443 nh->fib_nh_gw4 == gw &&
444 !(nh->fib_nh_flags & RTNH_F_DEAD)) {
445 spin_unlock(&fib_info_lock);
450 spin_unlock(&fib_info_lock);
455 static inline size_t fib_nlmsg_size(struct fib_info *fi)
457 size_t payload = NLMSG_ALIGN(sizeof(struct rtmsg))
458 + nla_total_size(4) /* RTA_TABLE */
459 + nla_total_size(4) /* RTA_DST */
460 + nla_total_size(4) /* RTA_PRIORITY */
461 + nla_total_size(4) /* RTA_PREFSRC */
462 + nla_total_size(TCP_CA_NAME_MAX); /* RTAX_CC_ALGO */
463 unsigned int nhs = fib_info_num_path(fi);
465 /* space for nested metrics */
466 payload += nla_total_size((RTAX_MAX * nla_total_size(4)));
469 payload += nla_total_size(4); /* RTA_NH_ID */
472 size_t nh_encapsize = 0;
473 /* Also handles the special case nhs == 1 */
475 /* each nexthop is packed in an attribute */
476 size_t nhsize = nla_total_size(sizeof(struct rtnexthop));
479 /* may contain flow and gateway attribute */
480 nhsize += 2 * nla_total_size(4);
482 /* grab encap info */
483 for (i = 0; i < fib_info_num_path(fi); i++) {
484 struct fib_nh_common *nhc = fib_info_nhc(fi, i);
486 if (nhc->nhc_lwtstate) {
488 nh_encapsize += lwtunnel_get_encap_size(
491 nh_encapsize += nla_total_size(2);
495 /* all nexthops are packed in a nested attribute */
496 payload += nla_total_size((nhs * nhsize) + nh_encapsize);
503 void rtmsg_fib(int event, __be32 key, struct fib_alias *fa,
504 int dst_len, u32 tb_id, const struct nl_info *info,
505 unsigned int nlm_flags)
507 struct fib_rt_info fri;
509 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0;
512 skb = nlmsg_new(fib_nlmsg_size(fa->fa_info), GFP_KERNEL);
516 fri.fi = fa->fa_info;
519 fri.dst_len = dst_len;
520 fri.tos = fa->fa_tos;
521 fri.type = fa->fa_type;
522 err = fib_dump_info(skb, info->portid, seq, event, &fri, nlm_flags);
524 /* -EMSGSIZE implies BUG in fib_nlmsg_size() */
525 WARN_ON(err == -EMSGSIZE);
529 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_IPV4_ROUTE,
530 info->nlh, GFP_KERNEL);
534 rtnl_set_sk_err(info->nl_net, RTNLGRP_IPV4_ROUTE, err);
537 static int fib_detect_death(struct fib_info *fi, int order,
538 struct fib_info **last_resort, int *last_idx,
541 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
543 int state = NUD_NONE;
545 if (likely(nhc->nhc_gw_family == AF_INET))
546 n = neigh_lookup(&arp_tbl, &nhc->nhc_gw.ipv4, nhc->nhc_dev);
547 else if (nhc->nhc_gw_family == AF_INET6)
548 n = neigh_lookup(ipv6_stub->nd_tbl, &nhc->nhc_gw.ipv6,
554 state = n->nud_state;
559 if (state == NUD_REACHABLE)
561 if ((state & NUD_VALID) && order != dflt)
563 if ((state & NUD_VALID) ||
564 (*last_idx < 0 && order > dflt && state != NUD_INCOMPLETE)) {
571 int fib_nh_common_init(struct fib_nh_common *nhc, struct nlattr *encap,
572 u16 encap_type, void *cfg, gfp_t gfp_flags,
573 struct netlink_ext_ack *extack)
577 nhc->nhc_pcpu_rth_output = alloc_percpu_gfp(struct rtable __rcu *,
579 if (!nhc->nhc_pcpu_rth_output)
583 struct lwtunnel_state *lwtstate;
585 if (encap_type == LWTUNNEL_ENCAP_NONE) {
586 NL_SET_ERR_MSG(extack, "LWT encap type not specified");
590 err = lwtunnel_build_state(encap_type, encap, nhc->nhc_family,
591 cfg, &lwtstate, extack);
595 nhc->nhc_lwtstate = lwtstate_get(lwtstate);
601 rt_fibinfo_free_cpus(nhc->nhc_pcpu_rth_output);
602 nhc->nhc_pcpu_rth_output = NULL;
605 EXPORT_SYMBOL_GPL(fib_nh_common_init);
607 int fib_nh_init(struct net *net, struct fib_nh *nh,
608 struct fib_config *cfg, int nh_weight,
609 struct netlink_ext_ack *extack)
613 nh->fib_nh_family = AF_INET;
615 err = fib_nh_common_init(&nh->nh_common, cfg->fc_encap,
616 cfg->fc_encap_type, cfg, GFP_KERNEL, extack);
620 nh->fib_nh_oif = cfg->fc_oif;
621 nh->fib_nh_gw_family = cfg->fc_gw_family;
622 if (cfg->fc_gw_family == AF_INET)
623 nh->fib_nh_gw4 = cfg->fc_gw4;
624 else if (cfg->fc_gw_family == AF_INET6)
625 nh->fib_nh_gw6 = cfg->fc_gw6;
627 nh->fib_nh_flags = cfg->fc_flags;
629 #ifdef CONFIG_IP_ROUTE_CLASSID
630 nh->nh_tclassid = cfg->fc_flow;
632 net->ipv4.fib_num_tclassid_users++;
634 #ifdef CONFIG_IP_ROUTE_MULTIPATH
635 nh->fib_nh_weight = nh_weight;
640 #ifdef CONFIG_IP_ROUTE_MULTIPATH
642 static int fib_count_nexthops(struct rtnexthop *rtnh, int remaining,
643 struct netlink_ext_ack *extack)
647 while (rtnh_ok(rtnh, remaining)) {
649 rtnh = rtnh_next(rtnh, &remaining);
652 /* leftover implies invalid nexthop configuration, discard it */
654 NL_SET_ERR_MSG(extack,
655 "Invalid nexthop configuration - extra data after nexthops");
662 /* only called when fib_nh is integrated into fib_info */
663 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
664 int remaining, struct fib_config *cfg,
665 struct netlink_ext_ack *extack)
667 struct net *net = fi->fib_net;
668 struct fib_config fib_cfg;
672 change_nexthops(fi) {
675 memset(&fib_cfg, 0, sizeof(fib_cfg));
677 if (!rtnh_ok(rtnh, remaining)) {
678 NL_SET_ERR_MSG(extack,
679 "Invalid nexthop configuration - extra data after nexthop");
683 if (rtnh->rtnh_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
684 NL_SET_ERR_MSG(extack,
685 "Invalid flags for nexthop - can not contain DEAD or LINKDOWN");
689 fib_cfg.fc_flags = (cfg->fc_flags & ~0xFF) | rtnh->rtnh_flags;
690 fib_cfg.fc_oif = rtnh->rtnh_ifindex;
692 attrlen = rtnh_attrlen(rtnh);
694 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
696 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
697 nlav = nla_find(attrs, attrlen, RTA_VIA);
699 NL_SET_ERR_MSG(extack,
700 "Nexthop configuration can not contain both GATEWAY and VIA");
704 fib_cfg.fc_gw4 = nla_get_in_addr(nla);
706 fib_cfg.fc_gw_family = AF_INET;
708 ret = fib_gw_from_via(&fib_cfg, nlav, extack);
713 nla = nla_find(attrs, attrlen, RTA_FLOW);
715 fib_cfg.fc_flow = nla_get_u32(nla);
717 fib_cfg.fc_encap = nla_find(attrs, attrlen, RTA_ENCAP);
718 nla = nla_find(attrs, attrlen, RTA_ENCAP_TYPE);
720 fib_cfg.fc_encap_type = nla_get_u16(nla);
723 ret = fib_nh_init(net, nexthop_nh, &fib_cfg,
724 rtnh->rtnh_hops + 1, extack);
728 rtnh = rtnh_next(rtnh, &remaining);
729 } endfor_nexthops(fi);
732 nh = fib_info_nh(fi, 0);
733 if (cfg->fc_oif && nh->fib_nh_oif != cfg->fc_oif) {
734 NL_SET_ERR_MSG(extack,
735 "Nexthop device index does not match RTA_OIF");
738 if (cfg->fc_gw_family) {
739 if (cfg->fc_gw_family != nh->fib_nh_gw_family ||
740 (cfg->fc_gw_family == AF_INET &&
741 nh->fib_nh_gw4 != cfg->fc_gw4) ||
742 (cfg->fc_gw_family == AF_INET6 &&
743 ipv6_addr_cmp(&nh->fib_nh_gw6, &cfg->fc_gw6))) {
744 NL_SET_ERR_MSG(extack,
745 "Nexthop gateway does not match RTA_GATEWAY or RTA_VIA");
749 #ifdef CONFIG_IP_ROUTE_CLASSID
750 if (cfg->fc_flow && nh->nh_tclassid != cfg->fc_flow) {
751 NL_SET_ERR_MSG(extack,
752 "Nexthop class id does not match RTA_FLOW");
761 /* only called when fib_nh is integrated into fib_info */
762 static void fib_rebalance(struct fib_info *fi)
767 if (fib_info_num_path(fi) < 2)
772 if (nh->fib_nh_flags & RTNH_F_DEAD)
775 if (ip_ignore_linkdown(nh->fib_nh_dev) &&
776 nh->fib_nh_flags & RTNH_F_LINKDOWN)
779 total += nh->fib_nh_weight;
780 } endfor_nexthops(fi);
783 change_nexthops(fi) {
786 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD) {
788 } else if (ip_ignore_linkdown(nexthop_nh->fib_nh_dev) &&
789 nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN) {
792 w += nexthop_nh->fib_nh_weight;
793 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31,
797 atomic_set(&nexthop_nh->fib_nh_upper_bound, upper_bound);
798 } endfor_nexthops(fi);
800 #else /* CONFIG_IP_ROUTE_MULTIPATH */
802 static int fib_get_nhs(struct fib_info *fi, struct rtnexthop *rtnh,
803 int remaining, struct fib_config *cfg,
804 struct netlink_ext_ack *extack)
806 NL_SET_ERR_MSG(extack, "Multipath support not enabled in kernel");
811 #define fib_rebalance(fi) do { } while (0)
813 #endif /* CONFIG_IP_ROUTE_MULTIPATH */
815 static int fib_encap_match(u16 encap_type,
816 struct nlattr *encap,
817 const struct fib_nh *nh,
818 const struct fib_config *cfg,
819 struct netlink_ext_ack *extack)
821 struct lwtunnel_state *lwtstate;
824 if (encap_type == LWTUNNEL_ENCAP_NONE)
827 ret = lwtunnel_build_state(encap_type, encap, AF_INET,
828 cfg, &lwtstate, extack);
830 result = lwtunnel_cmp_encap(lwtstate, nh->fib_nh_lws);
831 lwtstate_free(lwtstate);
837 int fib_nh_match(struct fib_config *cfg, struct fib_info *fi,
838 struct netlink_ext_ack *extack)
840 #ifdef CONFIG_IP_ROUTE_MULTIPATH
841 struct rtnexthop *rtnh;
845 if (cfg->fc_priority && cfg->fc_priority != fi->fib_priority)
849 if (fi->nh && cfg->fc_nh_id == fi->nh->id)
854 if (cfg->fc_oif || cfg->fc_gw_family) {
855 struct fib_nh *nh = fib_info_nh(fi, 0);
858 if (fib_encap_match(cfg->fc_encap_type, cfg->fc_encap,
862 #ifdef CONFIG_IP_ROUTE_CLASSID
864 cfg->fc_flow != nh->nh_tclassid)
867 if ((cfg->fc_oif && cfg->fc_oif != nh->fib_nh_oif) ||
868 (cfg->fc_gw_family &&
869 cfg->fc_gw_family != nh->fib_nh_gw_family))
872 if (cfg->fc_gw_family == AF_INET &&
873 cfg->fc_gw4 != nh->fib_nh_gw4)
876 if (cfg->fc_gw_family == AF_INET6 &&
877 ipv6_addr_cmp(&cfg->fc_gw6, &nh->fib_nh_gw6))
883 #ifdef CONFIG_IP_ROUTE_MULTIPATH
888 remaining = cfg->fc_mp_len;
893 if (!rtnh_ok(rtnh, remaining))
896 if (rtnh->rtnh_ifindex && rtnh->rtnh_ifindex != nh->fib_nh_oif)
899 attrlen = rtnh_attrlen(rtnh);
901 struct nlattr *nla, *nlav, *attrs = rtnh_attrs(rtnh);
903 nla = nla_find(attrs, attrlen, RTA_GATEWAY);
904 nlav = nla_find(attrs, attrlen, RTA_VIA);
906 NL_SET_ERR_MSG(extack,
907 "Nexthop configuration can not contain both GATEWAY and VIA");
912 if (nh->fib_nh_gw_family != AF_INET ||
913 nla_get_in_addr(nla) != nh->fib_nh_gw4)
916 struct fib_config cfg2;
919 err = fib_gw_from_via(&cfg2, nlav, extack);
923 switch (nh->fib_nh_gw_family) {
925 if (cfg2.fc_gw_family != AF_INET ||
926 cfg2.fc_gw4 != nh->fib_nh_gw4)
930 if (cfg2.fc_gw_family != AF_INET6 ||
931 ipv6_addr_cmp(&cfg2.fc_gw6,
938 #ifdef CONFIG_IP_ROUTE_CLASSID
939 nla = nla_find(attrs, attrlen, RTA_FLOW);
940 if (nla && nla_get_u32(nla) != nh->nh_tclassid)
945 rtnh = rtnh_next(rtnh, &remaining);
946 } endfor_nexthops(fi);
951 bool fib_metrics_match(struct fib_config *cfg, struct fib_info *fi)
959 nla_for_each_attr(nla, cfg->fc_mx, cfg->fc_mx_len, remaining) {
960 int type = nla_type(nla);
968 if (type == RTAX_CC_ALGO) {
969 char tmp[TCP_CA_NAME_MAX];
972 nla_strlcpy(tmp, nla, sizeof(tmp));
973 val = tcp_ca_get_key_by_name(fi->fib_net, tmp, &ecn_ca);
975 if (nla_len(nla) != sizeof(u32))
977 val = nla_get_u32(nla);
980 fi_val = fi->fib_metrics->metrics[type - 1];
981 if (type == RTAX_FEATURES)
982 fi_val &= ~DST_FEATURE_ECN_CA;
991 static int fib_check_nh_v6_gw(struct net *net, struct fib_nh *nh,
992 u32 table, struct netlink_ext_ack *extack)
994 struct fib6_config cfg = {
996 .fc_flags = nh->fib_nh_flags | RTF_GATEWAY,
997 .fc_ifindex = nh->fib_nh_oif,
998 .fc_gateway = nh->fib_nh_gw6,
1000 struct fib6_nh fib6_nh = {};
1003 err = ipv6_stub->fib6_nh_init(net, &fib6_nh, &cfg, GFP_KERNEL, extack);
1005 nh->fib_nh_dev = fib6_nh.fib_nh_dev;
1006 dev_hold(nh->fib_nh_dev);
1007 nh->fib_nh_oif = nh->fib_nh_dev->ifindex;
1008 nh->fib_nh_scope = RT_SCOPE_LINK;
1010 ipv6_stub->fib6_nh_release(&fib6_nh);
1020 * Semantics of nexthop is very messy by historical reasons.
1021 * We have to take into account, that:
1022 * a) gateway can be actually local interface address,
1023 * so that gatewayed route is direct.
1024 * b) gateway must be on-link address, possibly
1025 * described not by an ifaddr, but also by a direct route.
1026 * c) If both gateway and interface are specified, they should not
1028 * d) If we use tunnel routes, gateway could be not on-link.
1030 * Attempt to reconcile all of these (alas, self-contradictory) conditions
1031 * results in pretty ugly and hairy code with obscure logic.
1033 * I chose to generalized it instead, so that the size
1034 * of code does not increase practically, but it becomes
1035 * much more general.
1036 * Every prefix is assigned a "scope" value: "host" is local address,
1037 * "link" is direct route,
1038 * [ ... "site" ... "interior" ... ]
1039 * and "universe" is true gateway route with global meaning.
1041 * Every prefix refers to a set of "nexthop"s (gw, oif),
1042 * where gw must have narrower scope. This recursion stops
1043 * when gw has LOCAL scope or if "nexthop" is declared ONLINK,
1044 * which means that gw is forced to be on link.
1046 * Code is still hairy, but now it is apparently logically
1047 * consistent and very flexible. F.e. as by-product it allows
1048 * to co-exists in peace independent exterior and interior
1049 * routing processes.
1051 * Normally it looks as following.
1053 * {universe prefix} -> (gw, oif) [scope link]
1055 * |-> {link prefix} -> (gw, oif) [scope local]
1057 * |-> {local prefix} (terminal node)
1059 static int fib_check_nh_v4_gw(struct net *net, struct fib_nh *nh, u32 table,
1060 u8 scope, struct netlink_ext_ack *extack)
1062 struct net_device *dev;
1063 struct fib_result res;
1066 if (nh->fib_nh_flags & RTNH_F_ONLINK) {
1067 unsigned int addr_type;
1069 if (scope >= RT_SCOPE_LINK) {
1070 NL_SET_ERR_MSG(extack, "Nexthop has invalid scope");
1073 dev = __dev_get_by_index(net, nh->fib_nh_oif);
1075 NL_SET_ERR_MSG(extack, "Nexthop device required for onlink");
1078 if (!(dev->flags & IFF_UP)) {
1079 NL_SET_ERR_MSG(extack, "Nexthop device is not up");
1082 addr_type = inet_addr_type_dev_table(net, dev, nh->fib_nh_gw4);
1083 if (addr_type != RTN_UNICAST) {
1084 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1087 if (!netif_carrier_ok(dev))
1088 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1089 nh->fib_nh_dev = dev;
1091 nh->fib_nh_scope = RT_SCOPE_LINK;
1096 struct fib_table *tbl = NULL;
1097 struct flowi4 fl4 = {
1098 .daddr = nh->fib_nh_gw4,
1099 .flowi4_scope = scope + 1,
1100 .flowi4_oif = nh->fib_nh_oif,
1101 .flowi4_iif = LOOPBACK_IFINDEX,
1104 /* It is not necessary, but requires a bit of thinking */
1105 if (fl4.flowi4_scope < RT_SCOPE_LINK)
1106 fl4.flowi4_scope = RT_SCOPE_LINK;
1109 tbl = fib_get_table(net, table);
1112 err = fib_table_lookup(tbl, &fl4, &res,
1113 FIB_LOOKUP_IGNORE_LINKSTATE |
1116 /* on error or if no table given do full lookup. This
1117 * is needed for example when nexthops are in the local
1118 * table rather than the given table
1121 err = fib_lookup(net, &fl4, &res,
1122 FIB_LOOKUP_IGNORE_LINKSTATE);
1126 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1132 if (res.type != RTN_UNICAST && res.type != RTN_LOCAL) {
1133 NL_SET_ERR_MSG(extack, "Nexthop has invalid gateway");
1136 nh->fib_nh_scope = res.scope;
1137 nh->fib_nh_oif = FIB_RES_OIF(res);
1138 nh->fib_nh_dev = dev = FIB_RES_DEV(res);
1140 NL_SET_ERR_MSG(extack,
1141 "No egress device for nexthop gateway");
1145 if (!netif_carrier_ok(dev))
1146 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1147 err = (dev->flags & IFF_UP) ? 0 : -ENETDOWN;
1153 static int fib_check_nh_nongw(struct net *net, struct fib_nh *nh,
1154 struct netlink_ext_ack *extack)
1156 struct in_device *in_dev;
1159 if (nh->fib_nh_flags & (RTNH_F_PERVASIVE | RTNH_F_ONLINK)) {
1160 NL_SET_ERR_MSG(extack,
1161 "Invalid flags for nexthop - PERVASIVE and ONLINK can not be set");
1168 in_dev = inetdev_by_index(net, nh->fib_nh_oif);
1172 if (!(in_dev->dev->flags & IFF_UP)) {
1173 NL_SET_ERR_MSG(extack, "Device for nexthop is not up");
1177 nh->fib_nh_dev = in_dev->dev;
1178 dev_hold(nh->fib_nh_dev);
1179 nh->fib_nh_scope = RT_SCOPE_HOST;
1180 if (!netif_carrier_ok(nh->fib_nh_dev))
1181 nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1188 int fib_check_nh(struct net *net, struct fib_nh *nh, u32 table, u8 scope,
1189 struct netlink_ext_ack *extack)
1193 if (nh->fib_nh_gw_family == AF_INET)
1194 err = fib_check_nh_v4_gw(net, nh, table, scope, extack);
1195 else if (nh->fib_nh_gw_family == AF_INET6)
1196 err = fib_check_nh_v6_gw(net, nh, table, extack);
1198 err = fib_check_nh_nongw(net, nh, extack);
1203 static inline unsigned int fib_laddr_hashfn(__be32 val)
1205 unsigned int mask = (fib_info_hash_size - 1);
1207 return ((__force u32)val ^
1208 ((__force u32)val >> 7) ^
1209 ((__force u32)val >> 14)) & mask;
1212 static struct hlist_head *fib_info_hash_alloc(int bytes)
1214 if (bytes <= PAGE_SIZE)
1215 return kzalloc(bytes, GFP_KERNEL);
1217 return (struct hlist_head *)
1218 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
1222 static void fib_info_hash_free(struct hlist_head *hash, int bytes)
1227 if (bytes <= PAGE_SIZE)
1230 free_pages((unsigned long) hash, get_order(bytes));
1233 static void fib_info_hash_move(struct hlist_head *new_info_hash,
1234 struct hlist_head *new_laddrhash,
1235 unsigned int new_size)
1237 struct hlist_head *old_info_hash, *old_laddrhash;
1238 unsigned int old_size = fib_info_hash_size;
1239 unsigned int i, bytes;
1241 spin_lock_bh(&fib_info_lock);
1242 old_info_hash = fib_info_hash;
1243 old_laddrhash = fib_info_laddrhash;
1244 fib_info_hash_size = new_size;
1246 for (i = 0; i < old_size; i++) {
1247 struct hlist_head *head = &fib_info_hash[i];
1248 struct hlist_node *n;
1249 struct fib_info *fi;
1251 hlist_for_each_entry_safe(fi, n, head, fib_hash) {
1252 struct hlist_head *dest;
1253 unsigned int new_hash;
1255 new_hash = fib_info_hashfn(fi);
1256 dest = &new_info_hash[new_hash];
1257 hlist_add_head(&fi->fib_hash, dest);
1260 fib_info_hash = new_info_hash;
1262 for (i = 0; i < old_size; i++) {
1263 struct hlist_head *lhead = &fib_info_laddrhash[i];
1264 struct hlist_node *n;
1265 struct fib_info *fi;
1267 hlist_for_each_entry_safe(fi, n, lhead, fib_lhash) {
1268 struct hlist_head *ldest;
1269 unsigned int new_hash;
1271 new_hash = fib_laddr_hashfn(fi->fib_prefsrc);
1272 ldest = &new_laddrhash[new_hash];
1273 hlist_add_head(&fi->fib_lhash, ldest);
1276 fib_info_laddrhash = new_laddrhash;
1278 spin_unlock_bh(&fib_info_lock);
1280 bytes = old_size * sizeof(struct hlist_head *);
1281 fib_info_hash_free(old_info_hash, bytes);
1282 fib_info_hash_free(old_laddrhash, bytes);
1285 __be32 fib_info_update_nhc_saddr(struct net *net, struct fib_nh_common *nhc,
1286 unsigned char scope)
1290 if (nhc->nhc_family != AF_INET)
1291 return inet_select_addr(nhc->nhc_dev, 0, scope);
1293 nh = container_of(nhc, struct fib_nh, nh_common);
1294 nh->nh_saddr = inet_select_addr(nh->fib_nh_dev, nh->fib_nh_gw4, scope);
1295 nh->nh_saddr_genid = atomic_read(&net->ipv4.dev_addr_genid);
1297 return nh->nh_saddr;
1300 __be32 fib_result_prefsrc(struct net *net, struct fib_result *res)
1302 struct fib_nh_common *nhc = res->nhc;
1304 if (res->fi->fib_prefsrc)
1305 return res->fi->fib_prefsrc;
1307 if (nhc->nhc_family == AF_INET) {
1310 nh = container_of(nhc, struct fib_nh, nh_common);
1311 if (nh->nh_saddr_genid == atomic_read(&net->ipv4.dev_addr_genid))
1312 return nh->nh_saddr;
1315 return fib_info_update_nhc_saddr(net, nhc, res->fi->fib_scope);
1318 static bool fib_valid_prefsrc(struct fib_config *cfg, __be32 fib_prefsrc)
1320 if (cfg->fc_type != RTN_LOCAL || !cfg->fc_dst ||
1321 fib_prefsrc != cfg->fc_dst) {
1322 u32 tb_id = cfg->fc_table;
1325 if (tb_id == RT_TABLE_MAIN)
1326 tb_id = RT_TABLE_LOCAL;
1328 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1329 fib_prefsrc, tb_id);
1331 if (rc != RTN_LOCAL && tb_id != RT_TABLE_LOCAL) {
1332 rc = inet_addr_type_table(cfg->fc_nlinfo.nl_net,
1333 fib_prefsrc, RT_TABLE_LOCAL);
1336 if (rc != RTN_LOCAL)
1342 struct fib_info *fib_create_info(struct fib_config *cfg,
1343 struct netlink_ext_ack *extack)
1346 struct fib_info *fi = NULL;
1347 struct nexthop *nh = NULL;
1348 struct fib_info *ofi;
1350 struct net *net = cfg->fc_nlinfo.nl_net;
1352 if (cfg->fc_type > RTN_MAX)
1355 /* Fast check to catch the most weird cases */
1356 if (fib_props[cfg->fc_type].scope > cfg->fc_scope) {
1357 NL_SET_ERR_MSG(extack, "Invalid scope");
1361 if (cfg->fc_flags & (RTNH_F_DEAD | RTNH_F_LINKDOWN)) {
1362 NL_SET_ERR_MSG(extack,
1363 "Invalid rtm_flags - can not contain DEAD or LINKDOWN");
1367 if (cfg->fc_nh_id) {
1369 fi = fib_find_info_nh(net, cfg);
1376 nh = nexthop_find_by_id(net, cfg->fc_nh_id);
1378 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
1384 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1386 nhs = fib_count_nexthops(cfg->fc_mp, cfg->fc_mp_len, extack);
1393 if (fib_info_cnt >= fib_info_hash_size) {
1394 unsigned int new_size = fib_info_hash_size << 1;
1395 struct hlist_head *new_info_hash;
1396 struct hlist_head *new_laddrhash;
1401 bytes = new_size * sizeof(struct hlist_head *);
1402 new_info_hash = fib_info_hash_alloc(bytes);
1403 new_laddrhash = fib_info_hash_alloc(bytes);
1404 if (!new_info_hash || !new_laddrhash) {
1405 fib_info_hash_free(new_info_hash, bytes);
1406 fib_info_hash_free(new_laddrhash, bytes);
1408 fib_info_hash_move(new_info_hash, new_laddrhash, new_size);
1410 if (!fib_info_hash_size)
1414 fi = kzalloc(struct_size(fi, fib_nh, nhs), GFP_KERNEL);
1417 fi->fib_metrics = ip_fib_metrics_init(fi->fib_net, cfg->fc_mx,
1418 cfg->fc_mx_len, extack);
1419 if (IS_ERR(fi->fib_metrics)) {
1420 err = PTR_ERR(fi->fib_metrics);
1422 return ERR_PTR(err);
1427 fi->fib_protocol = cfg->fc_protocol;
1428 fi->fib_scope = cfg->fc_scope;
1429 fi->fib_flags = cfg->fc_flags;
1430 fi->fib_priority = cfg->fc_priority;
1431 fi->fib_prefsrc = cfg->fc_prefsrc;
1432 fi->fib_type = cfg->fc_type;
1433 fi->fib_tb_id = cfg->fc_table;
1437 if (!nexthop_get(nh)) {
1438 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
1445 change_nexthops(fi) {
1446 nexthop_nh->nh_parent = fi;
1447 } endfor_nexthops(fi)
1450 err = fib_get_nhs(fi, cfg->fc_mp, cfg->fc_mp_len, cfg,
1453 err = fib_nh_init(net, fi->fib_nh, cfg, 1, extack);
1459 if (fib_props[cfg->fc_type].error) {
1460 if (cfg->fc_gw_family || cfg->fc_oif || cfg->fc_mp) {
1461 NL_SET_ERR_MSG(extack,
1462 "Gateway, device and multipath can not be specified for this route type");
1467 switch (cfg->fc_type) {
1475 NL_SET_ERR_MSG(extack, "Invalid route type");
1480 if (cfg->fc_scope > RT_SCOPE_HOST) {
1481 NL_SET_ERR_MSG(extack, "Invalid scope");
1486 err = fib_check_nexthop(fi->nh, cfg->fc_scope, extack);
1489 } else if (cfg->fc_scope == RT_SCOPE_HOST) {
1490 struct fib_nh *nh = fi->fib_nh;
1492 /* Local address is added. */
1494 NL_SET_ERR_MSG(extack,
1495 "Route with host scope can not have multiple nexthops");
1498 if (nh->fib_nh_gw_family) {
1499 NL_SET_ERR_MSG(extack,
1500 "Route with host scope can not have a gateway");
1503 nh->fib_nh_scope = RT_SCOPE_NOWHERE;
1504 nh->fib_nh_dev = dev_get_by_index(net, nh->fib_nh_oif);
1506 if (!nh->fib_nh_dev)
1511 change_nexthops(fi) {
1512 err = fib_check_nh(cfg->fc_nlinfo.nl_net, nexthop_nh,
1513 cfg->fc_table, cfg->fc_scope,
1517 if (nexthop_nh->fib_nh_flags & RTNH_F_LINKDOWN)
1519 } endfor_nexthops(fi)
1520 if (linkdown == fi->fib_nhs)
1521 fi->fib_flags |= RTNH_F_LINKDOWN;
1524 if (fi->fib_prefsrc && !fib_valid_prefsrc(cfg, fi->fib_prefsrc)) {
1525 NL_SET_ERR_MSG(extack, "Invalid prefsrc address");
1530 change_nexthops(fi) {
1531 fib_info_update_nhc_saddr(net, &nexthop_nh->nh_common,
1533 if (nexthop_nh->fib_nh_gw_family == AF_INET6)
1534 fi->fib_nh_is_v6 = true;
1535 } endfor_nexthops(fi)
1541 ofi = fib_find_info(fi);
1550 refcount_set(&fi->fib_clntref, 1);
1551 spin_lock_bh(&fib_info_lock);
1552 hlist_add_head(&fi->fib_hash,
1553 &fib_info_hash[fib_info_hashfn(fi)]);
1554 if (fi->fib_prefsrc) {
1555 struct hlist_head *head;
1557 head = &fib_info_laddrhash[fib_laddr_hashfn(fi->fib_prefsrc)];
1558 hlist_add_head(&fi->fib_lhash, head);
1561 list_add(&fi->nh_list, &nh->fi_list);
1563 change_nexthops(fi) {
1564 struct hlist_head *head;
1567 if (!nexthop_nh->fib_nh_dev)
1569 hash = fib_devindex_hashfn(nexthop_nh->fib_nh_dev->ifindex);
1570 head = &fib_info_devhash[hash];
1571 hlist_add_head(&nexthop_nh->nh_hash, head);
1572 } endfor_nexthops(fi)
1574 spin_unlock_bh(&fib_info_lock);
1586 return ERR_PTR(err);
1589 int fib_nexthop_info(struct sk_buff *skb, const struct fib_nh_common *nhc,
1590 u8 rt_family, unsigned char *flags, bool skip_oif)
1592 if (nhc->nhc_flags & RTNH_F_DEAD)
1593 *flags |= RTNH_F_DEAD;
1595 if (nhc->nhc_flags & RTNH_F_LINKDOWN) {
1596 *flags |= RTNH_F_LINKDOWN;
1599 switch (nhc->nhc_family) {
1601 if (ip_ignore_linkdown(nhc->nhc_dev))
1602 *flags |= RTNH_F_DEAD;
1605 if (ip6_ignore_linkdown(nhc->nhc_dev))
1606 *flags |= RTNH_F_DEAD;
1612 switch (nhc->nhc_gw_family) {
1614 if (nla_put_in_addr(skb, RTA_GATEWAY, nhc->nhc_gw.ipv4))
1615 goto nla_put_failure;
1618 /* if gateway family does not match nexthop family
1619 * gateway is encoded as RTA_VIA
1621 if (rt_family != nhc->nhc_gw_family) {
1622 int alen = sizeof(struct in6_addr);
1626 nla = nla_reserve(skb, RTA_VIA, alen + 2);
1628 goto nla_put_failure;
1630 via = nla_data(nla);
1631 via->rtvia_family = AF_INET6;
1632 memcpy(via->rtvia_addr, &nhc->nhc_gw.ipv6, alen);
1633 } else if (nla_put_in6_addr(skb, RTA_GATEWAY,
1634 &nhc->nhc_gw.ipv6) < 0) {
1635 goto nla_put_failure;
1640 *flags |= (nhc->nhc_flags & RTNH_F_ONLINK);
1641 if (nhc->nhc_flags & RTNH_F_OFFLOAD)
1642 *flags |= RTNH_F_OFFLOAD;
1644 if (!skip_oif && nhc->nhc_dev &&
1645 nla_put_u32(skb, RTA_OIF, nhc->nhc_dev->ifindex))
1646 goto nla_put_failure;
1648 if (nhc->nhc_lwtstate &&
1649 lwtunnel_fill_encap(skb, nhc->nhc_lwtstate,
1650 RTA_ENCAP, RTA_ENCAP_TYPE) < 0)
1651 goto nla_put_failure;
1658 EXPORT_SYMBOL_GPL(fib_nexthop_info);
1660 #if IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) || IS_ENABLED(CONFIG_IPV6)
1661 int fib_add_nexthop(struct sk_buff *skb, const struct fib_nh_common *nhc,
1662 int nh_weight, u8 rt_family)
1664 const struct net_device *dev = nhc->nhc_dev;
1665 struct rtnexthop *rtnh;
1666 unsigned char flags = 0;
1668 rtnh = nla_reserve_nohdr(skb, sizeof(*rtnh));
1670 goto nla_put_failure;
1672 rtnh->rtnh_hops = nh_weight - 1;
1673 rtnh->rtnh_ifindex = dev ? dev->ifindex : 0;
1675 if (fib_nexthop_info(skb, nhc, rt_family, &flags, true) < 0)
1676 goto nla_put_failure;
1678 rtnh->rtnh_flags = flags;
1680 /* length of rtnetlink header + attributes */
1681 rtnh->rtnh_len = nlmsg_get_pos(skb) - (void *)rtnh;
1688 EXPORT_SYMBOL_GPL(fib_add_nexthop);
1691 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1692 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1696 mp = nla_nest_start_noflag(skb, RTA_MULTIPATH);
1698 goto nla_put_failure;
1700 if (unlikely(fi->nh)) {
1701 if (nexthop_mpath_fill_node(skb, fi->nh, AF_INET) < 0)
1702 goto nla_put_failure;
1707 if (fib_add_nexthop(skb, &nh->nh_common, nh->fib_nh_weight,
1709 goto nla_put_failure;
1710 #ifdef CONFIG_IP_ROUTE_CLASSID
1711 if (nh->nh_tclassid &&
1712 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1713 goto nla_put_failure;
1715 } endfor_nexthops(fi);
1718 nla_nest_end(skb, mp);
1726 static int fib_add_multipath(struct sk_buff *skb, struct fib_info *fi)
1732 int fib_dump_info(struct sk_buff *skb, u32 portid, u32 seq, int event,
1733 struct fib_rt_info *fri, unsigned int flags)
1735 unsigned int nhs = fib_info_num_path(fri->fi);
1736 struct fib_info *fi = fri->fi;
1737 u32 tb_id = fri->tb_id;
1738 struct nlmsghdr *nlh;
1741 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
1745 rtm = nlmsg_data(nlh);
1746 rtm->rtm_family = AF_INET;
1747 rtm->rtm_dst_len = fri->dst_len;
1748 rtm->rtm_src_len = 0;
1749 rtm->rtm_tos = fri->tos;
1751 rtm->rtm_table = tb_id;
1753 rtm->rtm_table = RT_TABLE_COMPAT;
1754 if (nla_put_u32(skb, RTA_TABLE, tb_id))
1755 goto nla_put_failure;
1756 rtm->rtm_type = fri->type;
1757 rtm->rtm_flags = fi->fib_flags;
1758 rtm->rtm_scope = fi->fib_scope;
1759 rtm->rtm_protocol = fi->fib_protocol;
1761 if (rtm->rtm_dst_len &&
1762 nla_put_in_addr(skb, RTA_DST, fri->dst))
1763 goto nla_put_failure;
1764 if (fi->fib_priority &&
1765 nla_put_u32(skb, RTA_PRIORITY, fi->fib_priority))
1766 goto nla_put_failure;
1767 if (rtnetlink_put_metrics(skb, fi->fib_metrics->metrics) < 0)
1768 goto nla_put_failure;
1770 if (fi->fib_prefsrc &&
1771 nla_put_in_addr(skb, RTA_PREFSRC, fi->fib_prefsrc))
1772 goto nla_put_failure;
1775 if (nla_put_u32(skb, RTA_NH_ID, fi->nh->id))
1776 goto nla_put_failure;
1777 if (nexthop_is_blackhole(fi->nh))
1778 rtm->rtm_type = RTN_BLACKHOLE;
1782 const struct fib_nh_common *nhc = fib_info_nhc(fi, 0);
1783 unsigned char flags = 0;
1785 if (fib_nexthop_info(skb, nhc, AF_INET, &flags, false) < 0)
1786 goto nla_put_failure;
1788 rtm->rtm_flags = flags;
1789 #ifdef CONFIG_IP_ROUTE_CLASSID
1790 if (nhc->nhc_family == AF_INET) {
1793 nh = container_of(nhc, struct fib_nh, nh_common);
1794 if (nh->nh_tclassid &&
1795 nla_put_u32(skb, RTA_FLOW, nh->nh_tclassid))
1796 goto nla_put_failure;
1800 if (fib_add_multipath(skb, fi) < 0)
1801 goto nla_put_failure;
1804 nlmsg_end(skb, nlh);
1808 nlmsg_cancel(skb, nlh);
1814 * - local address disappeared -> we must delete all the entries
1816 * - device went down -> we must shutdown all nexthops going via it.
1818 int fib_sync_down_addr(struct net_device *dev, __be32 local)
1821 unsigned int hash = fib_laddr_hashfn(local);
1822 struct hlist_head *head = &fib_info_laddrhash[hash];
1823 int tb_id = l3mdev_fib_table(dev) ? : RT_TABLE_MAIN;
1824 struct net *net = dev_net(dev);
1825 struct fib_info *fi;
1827 if (!fib_info_laddrhash || local == 0)
1830 hlist_for_each_entry(fi, head, fib_lhash) {
1831 if (!net_eq(fi->fib_net, net) ||
1832 fi->fib_tb_id != tb_id)
1834 if (fi->fib_prefsrc == local) {
1835 fi->fib_flags |= RTNH_F_DEAD;
1842 static int call_fib_nh_notifiers(struct fib_nh *nh,
1843 enum fib_event_type event_type)
1845 bool ignore_link_down = ip_ignore_linkdown(nh->fib_nh_dev);
1846 struct fib_nh_notifier_info info = {
1850 switch (event_type) {
1851 case FIB_EVENT_NH_ADD:
1852 if (nh->fib_nh_flags & RTNH_F_DEAD)
1854 if (ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN)
1856 return call_fib4_notifiers(dev_net(nh->fib_nh_dev), event_type,
1858 case FIB_EVENT_NH_DEL:
1859 if ((ignore_link_down && nh->fib_nh_flags & RTNH_F_LINKDOWN) ||
1860 (nh->fib_nh_flags & RTNH_F_DEAD))
1861 return call_fib4_notifiers(dev_net(nh->fib_nh_dev),
1862 event_type, &info.info);
1870 /* Update the PMTU of exceptions when:
1871 * - the new MTU of the first hop becomes smaller than the PMTU
1872 * - the old MTU was the same as the PMTU, and it limited discovery of
1873 * larger MTUs on the path. With that limit raised, we can now
1874 * discover larger MTUs
1875 * A special case is locked exceptions, for which the PMTU is smaller
1876 * than the minimal accepted PMTU:
1877 * - if the new MTU is greater than the PMTU, don't make any change
1878 * - otherwise, unlock and set PMTU
1880 void fib_nhc_update_mtu(struct fib_nh_common *nhc, u32 new, u32 orig)
1882 struct fnhe_hash_bucket *bucket;
1885 bucket = rcu_dereference_protected(nhc->nhc_exceptions, 1);
1889 for (i = 0; i < FNHE_HASH_SIZE; i++) {
1890 struct fib_nh_exception *fnhe;
1892 for (fnhe = rcu_dereference_protected(bucket[i].chain, 1);
1894 fnhe = rcu_dereference_protected(fnhe->fnhe_next, 1)) {
1895 if (fnhe->fnhe_mtu_locked) {
1896 if (new <= fnhe->fnhe_pmtu) {
1897 fnhe->fnhe_pmtu = new;
1898 fnhe->fnhe_mtu_locked = false;
1900 } else if (new < fnhe->fnhe_pmtu ||
1901 orig == fnhe->fnhe_pmtu) {
1902 fnhe->fnhe_pmtu = new;
1908 void fib_sync_mtu(struct net_device *dev, u32 orig_mtu)
1910 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1911 struct hlist_head *head = &fib_info_devhash[hash];
1914 hlist_for_each_entry(nh, head, nh_hash) {
1915 if (nh->fib_nh_dev == dev)
1916 fib_nhc_update_mtu(&nh->nh_common, dev->mtu, orig_mtu);
1920 /* Event force Flags Description
1921 * NETDEV_CHANGE 0 LINKDOWN Carrier OFF, not for scope host
1922 * NETDEV_DOWN 0 LINKDOWN|DEAD Link down, not for scope host
1923 * NETDEV_DOWN 1 LINKDOWN|DEAD Last address removed
1924 * NETDEV_UNREGISTER 1 LINKDOWN|DEAD Device removed
1926 * only used when fib_nh is built into fib_info
1928 int fib_sync_down_dev(struct net_device *dev, unsigned long event, bool force)
1931 int scope = RT_SCOPE_NOWHERE;
1932 struct fib_info *prev_fi = NULL;
1933 unsigned int hash = fib_devindex_hashfn(dev->ifindex);
1934 struct hlist_head *head = &fib_info_devhash[hash];
1940 hlist_for_each_entry(nh, head, nh_hash) {
1941 struct fib_info *fi = nh->nh_parent;
1944 BUG_ON(!fi->fib_nhs);
1945 if (nh->fib_nh_dev != dev || fi == prev_fi)
1949 change_nexthops(fi) {
1950 if (nexthop_nh->fib_nh_flags & RTNH_F_DEAD)
1952 else if (nexthop_nh->fib_nh_dev == dev &&
1953 nexthop_nh->fib_nh_scope != scope) {
1956 case NETDEV_UNREGISTER:
1957 nexthop_nh->fib_nh_flags |= RTNH_F_DEAD;
1960 nexthop_nh->fib_nh_flags |= RTNH_F_LINKDOWN;
1963 call_fib_nh_notifiers(nexthop_nh,
1967 #ifdef CONFIG_IP_ROUTE_MULTIPATH
1968 if (event == NETDEV_UNREGISTER &&
1969 nexthop_nh->fib_nh_dev == dev) {
1974 } endfor_nexthops(fi)
1975 if (dead == fi->fib_nhs) {
1978 case NETDEV_UNREGISTER:
1979 fi->fib_flags |= RTNH_F_DEAD;
1982 fi->fib_flags |= RTNH_F_LINKDOWN;
1994 /* Must be invoked inside of an RCU protected region. */
1995 static void fib_select_default(const struct flowi4 *flp, struct fib_result *res)
1997 struct fib_info *fi = NULL, *last_resort = NULL;
1998 struct hlist_head *fa_head = res->fa_head;
1999 struct fib_table *tb = res->table;
2000 u8 slen = 32 - res->prefixlen;
2001 int order = -1, last_idx = -1;
2002 struct fib_alias *fa, *fa1 = NULL;
2003 u32 last_prio = res->fi->fib_priority;
2006 hlist_for_each_entry_rcu(fa, fa_head, fa_list) {
2007 struct fib_info *next_fi = fa->fa_info;
2010 if (fa->fa_slen != slen)
2012 if (fa->fa_tos && fa->fa_tos != flp->flowi4_tos)
2014 if (fa->tb_id != tb->tb_id)
2016 if (next_fi->fib_priority > last_prio &&
2017 fa->fa_tos == last_tos) {
2022 if (next_fi->fib_flags & RTNH_F_DEAD)
2024 last_tos = fa->fa_tos;
2025 last_prio = next_fi->fib_priority;
2027 if (next_fi->fib_scope != res->scope ||
2028 fa->fa_type != RTN_UNICAST)
2031 nh = fib_info_nh(next_fi, 0);
2032 if (!nh->fib_nh_gw4 || nh->fib_nh_scope != RT_SCOPE_LINK)
2035 fib_alias_accessed(fa);
2038 if (next_fi != res->fi)
2041 } else if (!fib_detect_death(fi, order, &last_resort,
2042 &last_idx, fa1->fa_default)) {
2043 fib_result_assign(res, fi);
2044 fa1->fa_default = order;
2051 if (order <= 0 || !fi) {
2053 fa1->fa_default = -1;
2057 if (!fib_detect_death(fi, order, &last_resort, &last_idx,
2059 fib_result_assign(res, fi);
2060 fa1->fa_default = order;
2065 fib_result_assign(res, last_resort);
2066 fa1->fa_default = last_idx;
2072 * Dead device goes up. We wake up dead nexthops.
2073 * It takes sense only on multipath routes.
2075 * only used when fib_nh is built into fib_info
2077 int fib_sync_up(struct net_device *dev, unsigned char nh_flags)
2079 struct fib_info *prev_fi;
2081 struct hlist_head *head;
2085 if (!(dev->flags & IFF_UP))
2088 if (nh_flags & RTNH_F_DEAD) {
2089 unsigned int flags = dev_get_flags(dev);
2091 if (flags & (IFF_RUNNING | IFF_LOWER_UP))
2092 nh_flags |= RTNH_F_LINKDOWN;
2096 hash = fib_devindex_hashfn(dev->ifindex);
2097 head = &fib_info_devhash[hash];
2100 hlist_for_each_entry(nh, head, nh_hash) {
2101 struct fib_info *fi = nh->nh_parent;
2104 BUG_ON(!fi->fib_nhs);
2105 if (nh->fib_nh_dev != dev || fi == prev_fi)
2110 change_nexthops(fi) {
2111 if (!(nexthop_nh->fib_nh_flags & nh_flags)) {
2115 if (!nexthop_nh->fib_nh_dev ||
2116 !(nexthop_nh->fib_nh_dev->flags & IFF_UP))
2118 if (nexthop_nh->fib_nh_dev != dev ||
2119 !__in_dev_get_rtnl(dev))
2122 nexthop_nh->fib_nh_flags &= ~nh_flags;
2123 call_fib_nh_notifiers(nexthop_nh, FIB_EVENT_NH_ADD);
2124 } endfor_nexthops(fi)
2127 fi->fib_flags &= ~nh_flags;
2137 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2138 static bool fib_good_nh(const struct fib_nh *nh)
2140 int state = NUD_REACHABLE;
2142 if (nh->fib_nh_scope == RT_SCOPE_LINK) {
2143 struct neighbour *n;
2147 if (likely(nh->fib_nh_gw_family == AF_INET))
2148 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev,
2149 (__force u32)nh->fib_nh_gw4);
2150 else if (nh->fib_nh_gw_family == AF_INET6)
2151 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev,
2156 state = n->nud_state;
2158 rcu_read_unlock_bh();
2161 return !!(state & NUD_VALID);
2164 void fib_select_multipath(struct fib_result *res, int hash)
2166 struct fib_info *fi = res->fi;
2167 struct net *net = fi->fib_net;
2170 if (unlikely(res->fi->nh)) {
2171 nexthop_path_fib_result(res, hash);
2175 change_nexthops(fi) {
2176 if (net->ipv4.sysctl_fib_multipath_use_neigh) {
2177 if (!fib_good_nh(nexthop_nh))
2180 res->nh_sel = nhsel;
2181 res->nhc = &nexthop_nh->nh_common;
2186 if (hash > atomic_read(&nexthop_nh->fib_nh_upper_bound))
2189 res->nh_sel = nhsel;
2190 res->nhc = &nexthop_nh->nh_common;
2192 } endfor_nexthops(fi);
2196 void fib_select_path(struct net *net, struct fib_result *res,
2197 struct flowi4 *fl4, const struct sk_buff *skb)
2199 if (fl4->flowi4_oif && !(fl4->flowi4_flags & FLOWI_FLAG_SKIP_NH_OIF))
2202 #ifdef CONFIG_IP_ROUTE_MULTIPATH
2203 if (fib_info_num_path(res->fi) > 1) {
2204 int h = fib_multipath_hash(net, fl4, skb, NULL);
2206 fib_select_multipath(res, h);
2210 if (!res->prefixlen &&
2211 res->table->tb_num_default > 1 &&
2212 res->type == RTN_UNICAST)
2213 fib_select_default(fl4, res);
2217 fl4->saddr = fib_result_prefsrc(net, res);